CHOOSING & WORKING WITH AN AGENCY · September 2026 · ~10 min read
Signs your current agency has stopped doing the work
You can check this yourself in about twenty minutes, without asking anyone. Look at your sitemap for the date of the last published page, your profile for the date of the last post, and your last three reports side by side to see whether the wording changed more than the work. Output leaves a timestamp. Effort does not.
On this page
- 01What can I check myself without asking?
- 02How do I turn the gap into a number?
- 03Which reporting activity is theatre?
- 04Which changes in the reporting actually mean something?
- 05Is quiet always a bad sign?
- 06How do I raise it without blowing up the relationship?
- 07What to do this week
- 08When you do not need this
- 09Sources
- 10Related reading
- 11Questions about whether the work is still happening?
Drift is the normal failure mode, not theft. An account gets reassigned, a key person leaves, a bigger client absorbs the hours, and your account keeps getting the reporting without getting the production. Nobody decides this. It accumulates.
The reason to check with evidence rather than instinct is that your instinct will be wrong in both directions. Owners get suspicious during slow months when the work is fine, and stay comfortable during quiet quarters when it is not.
01What can I check myself without asking?
Four places, all yours to look at.
Your sitemap. Type your domain followed by /sitemap.xml and look at the dates. Some platforms publish an index file instead, and your robots.txt file names the real location if the default does not load. New and updated pages carry a timestamp. If the newest date is four months old and your scope says four pages a month, you have a factual gap rather than a feeling.
Your business profile. Open it and look at the posts, the photos, and the products or services list. Post history is visible with dates on it.
Search Console. The page indexing report shows how many of your pages are indexed and which are not, with reasons. Write today's count and today's date down, because that report does not hold a long history and a number you did not record is a comparison you cannot make later. The performance report is the one that keeps a longer window, so use it for trend and use the indexing report for a point in time.
Your reviews. Whether new ones are arriving, and whether responses are being written. Response text is public and dated.
None of this requires access to anything the agency holds. That is the point. Everything above is your own property, viewed from outside.
02How do I turn the gap into a number?
Worked example
Count the contracted units, count the delivered units, divide. It takes fifteen minutes and it converts an argument about effort into an arithmetic about output.
Step one, the content line. Say your scope is four pages a month. Over ninety days that is twelve contracted pages. Your sitemap shows five with dates inside the window. You received 42%derived of the contracted output, and the shortfall is 58%derived of that line.
Step two, price the shortfall against the fee. Ask what share of the monthly hours the content line represents. Say it is roughly a third. The missing work is then about 19%derived of everything you paid for in the quarter. That is the sentence to put in the email, because it is specific and it is checkable by both sides.
Step three, do the same on the line you can verify from outside. Reviews are the easiest. Count the reviews received in the window and count how many have owner responses. Twenty two reviews with nine responses is a 41%derived response rate against a scope that says every review within one business day.
Step four, check the direction of the review flow itself, because this is the one that costs you rankings quietly. The right metrics are reviews per month against your competitors' rate, and days since your last review. Total review count is a vanity number that hides a profile that stopped moving six months ago. Whitespark's 2026 survey of forty seven local search practitioners ranked review recency eleventh among local pack factors and a sustained flow rather than bursts fourteenth, and Darren Shaw's stated position is that when new reviews stop, rankings start to slip. Sterling Sky documented a client whose rankings recovered once the review flow resumed after flat lining.
Two honest limits on that. It is expert opinion and case observation rather than a controlled test, and the survey's own author says the panel has no access to the algorithm. And a slow review month can be your fault rather than the agency's, if the request system runs off a booking platform you stopped using.
Which reporting activity is theatre?
The review response line is the one to inspect first, because it is highly visible, easy to automate, and mostly does not do what it is sold as doing.
Responding to reviews is worth doing. The honest case for it is conversion and rating repair: Google notes that a reply can prompt a customer to update a review, and a one star converted to a five star is worth more than several new ones. What there is no controlled test for, in any source in this library, is that responding improves rankings. Whitespark's panel includes response behaviour inside its review signals category, but that is opinion, and there is at least one documented case where the same panel's view did not survive a controlled test.
Two things make a response line pure theatre, and both are checkable from outside. The first is keyword stuffing the replies. Keywords in owner responses to reviews scored 28 in the 2026 survey, close to the bottom of 187 factors, against 227 for the top factor. If your replies read like they were written for a search engine, you are paying for a low scoring tactic that makes your business sound strange to the person who left the review.
The second is automation with no human in the loop. Google's own help documentation tells businesses not to send the same thank you to everyone, and to focus on replies where they can share a useful update or answer a question. That is Google, in writing, describing what an unattended auto responder does as the thing not to do. Open your last twenty replies and read them in a row. If they are interchangeable, nobody is working the account, whatever the report says.
For a healthcare client this stops being a quality question and becomes a legal one, because a public reply that acknowledges someone was a patient is a disclosure. Never let an unattended tool post to a healthcare profile.
04Which changes in the reporting actually mean something?
Put three consecutive reports next to each other and read them like a P&L.
Look for the metrics changing rather than the numbers. When results are weak, reporting tends to drift toward whatever is still going up. Impressions replace calls. Keywords tracked replaces keywords ranking. Total sessions replaces converting sessions. That substitution is more informative than any single figure.
Look for the commentary getting shorter. A page of explanation in month one, three sentences by month eight.
Look for whether bad months get explained without you asking. An agency doing real work usually knows before you do why a number dropped and will say so.
And check whether the report still ties to the starting values you agreed. Numbers with no reference point are exactly the flaw that makes agency case studies unreadable, and a result presented without a baseline should be read the same way whoever it comes from.
One caution before you accuse anyone of hiding AI results. Google's generative AI performance report in Search Console publishes impressions by page, country, device and date, and explicitly does not publish clicks, click through rate, position, or the prompt a person typed. Clicks arriving from AI Overviews are also included in ordinary organic traffic with no separate label. An agency that tells you it cannot report AI clicks is telling you the truth. An agency that reports them confidently is the problem, and there is a straight answer a client deserves when they ask about AI.
05Is quiet always a bad sign?
No, and this is where owners get it wrong most often.
Some months are genuinely investigation. Technical diagnosis, competitive research, or a migration plan produces very little you can see until it lands. That is real work and it looks like nothing.
Some work is slow by nature. Local search results move over months, and what a first ninety days can honestly produce is less visible than most people expect. Confusing a slow channel with a lazy vendor is a common and expensive mistake.
Some work is legitimately unglamorous. Directory listings are the usual example, and citation work sits somewhere between necessary maintenance and theatre depending on how it is done. A report full of citations is not automatically padding, though it is worth asking which ones and why.
And sometimes the bottleneck is you. If approvals sit in your inbox for two weeks, production stops and the agency is often too polite to say it plainly. Check your own sent folder before you check theirs.
06How do I raise it without blowing up the relationship?
Ask for a change log, not an explanation.
The email is short. "Can you send me a dated list of everything published, changed, or submitted in the last ninety days?" That is a normal request, it is not accusatory, and it is answerable in an hour by anyone who did the work.
What comes back tells you what you need. A dated list means the work happened. A meeting invitation instead of a list, twice, means something else.
Then ask one forward looking question: what are the next three things and by when. Vagueness about the past can be a records problem. Vagueness about the next thirty days usually is not.
07What to do this week
Run the twenty minute check. Sitemap date, profile post date, indexed page count, review responses. Write the four dates on one line, with today's date next to them.
Run the arithmetic. Contracted units against delivered units, then the shortfall as a share of the fee.
Send the change log request. Give a specific window, ninety days, and ask for it in a list rather than a call.
Then ask the question that a good agency answers easily and a drifting one cannot, and it applies to us too. Which month in the last year did you under deliver on this scope, and what did you do about it? SEOD would have to answer that with a month and a remedy or admit we are not counting. In any twelve months of real production there is at least one month that came up short. An agency that has never reported one has either been perfect or is not measuring itself, and only one of those is likely.
Compare what arrives against your scope of work. Not against your mood, against the document you signed. If the two do not match, start with the gap rather than the grievance. A good agency will fix it. Some will admit the account slipped, which is worth more than a defense. If it happens twice, you know.
Be honest with yourself
When you do not need this
If you are in the first sixty days, stop looking. The work at that stage is setup and research, and demanding visible output now will push your agency toward doing shallow things quickly. It is the same reason a real discovery process feels slow while it is happening.
If your reports arrive on time with numbers tied to a baseline and explanations for the bad months, you are being served well. Do not manufacture a problem.
And if the actual issue is that results are disappointing while the work is genuinely being done, this is the wrong article. That is a strategy conversation, not an accountability one, and treating it as neglect will cost you a vendor who was doing their job.
Sources
- Whitespark, "Local Search Ranking Factors," Darren Shaw, 6 November 2025. Forty seven experts scoring 187 factors. Source of the review recency and sustained flow rankings and the score for keywords in owner responses. Vendor research, and expert opinion rather than test data.
- Google, "Reply to Google reviews" guidance for businesses. Google's own instruction not to send the same thank you to everyone. First party platform documentation.
- Sterling Sky, Joy Hawkins, "Does Review Recency Impact Ranking?". The documented case of a flat lined review flow and the recovery when it resumed. Practitioner case study, small sample.
- Whitespark, "Review Recency is the Most Underrated Local Ranking Factor in 2025," Darren Shaw, 2 May 2025. Source of the position that rankings slip when new reviews stop. Vendor research and practitioner opinion.
Related reading
- How often you should hear from your agency. The cadence that should have surfaced the gap before you had to go looking.
- What a marketing retainer should actually include. The scope document you are counting the delivered units against.
- Who owns your website when the relationship ends. What to secure before you raise any of this, if it is heading toward a parting.
- The claims agencies make that do not survive checking. How to trace the numbers in the reports you just compared.
Questions about whether the work is still happening?
Email me at eric@seod.com with your last three monthly reports, names redacted if you want, and I will tell you whether the work changed or only the wording. You will get back a short note on what the reports stopped measuring and what to ask for. I will not offer to replace your agency, and if the reports look healthy I will say that instead.
There is more on hiring and working with agencies whenever you need it.